Student name:Ibrahim Saad Mahmoud
Subject:operative
level:four
ARTIFICIAL INTELLIGENCE IN RESTORATIVE
DENTISTRY
• Introduction
Artificial intelligence (AI) is a field of computer science that engages in rendering
machines the ability to look like they have human intelligence or the capability to
replicate intelligent human behaviors. In other words, AI is intelligence that is shown
by machines, computers, and systems, whereas natural intelligence is represented
by animals and humans. Accordingly, any machine that recognizes its surroundings
and reacts to them for attaining specific objectives is considered as an intelligent
agent. In this context, it is not surprising that the use of AI is significantly increasing
in dentistry in which diagnosis, clinical decision-making, and prediction of treatment
outcomes play fundamental roles. AI systems can help dentists with designing
clinical decision support structures, and therefore, dentists should be aware of how
AI works and how they can be assisted by AI systems.
• Caries Detection
For preserving tooth tissues, delaying restorative interventions, diminishing
expensive restorative treatments, and retaining teeth long-term, early detection of
non-cavitated lesions is a must for controlling these lesions through noninvasive
treatments such as applying fluoride varnishes, infiltrating caries, and sealing
fissures. The diagnosis of non-cavitated caries lesion with direct visual and tactile
exploration may be inadequate for proximal tooth surfaces, and therefore, bitewing
radiography is additionally suggested. However, the results of radiographic
assessments for caries diagnosis may be affected by dental clinicians and examiners
due to their false-positive or false-negative detections.
• Application of AI for caries detection
AI models have been proposed as assistance to dental examiners when analyzing
radiographic and/or photographic images in order to improve caries diagnosis in
terms of accuracy and precision. AI models can employ intraoral digital images,
bitewing radiographs, or panoramic radiographs to recognize caries lesions and
offer appropriate treatments accordingly; however, panoramic radiographs may be
less viable when it comes to caries detection.
• Function of AI for caries detection
DL (deep learning) using CNNs(convolutional neural networks) has been mostly used
to develop an AI-based system for caries detection. CNNs are able to map images as
inputs and provide image identification and classification as outputs according to
the weights obtained from training data. To train the AI-based system, images are
initially labeled, and then, both the images and their labels are supplied to the CNN.
The CNN is computationally programmed so that it can diagnose the presence of a
labeled entity such as a caries lesion on unseen images. Indeed, CNNs are well-
designed for image classification and are the most commonly used algorithms for
caries identification.
• Accuracy of AI for caries detection
When assessing the effectiveness of ML, the diagnostic accuracy of a machine is
basically crucial and should be ideally 100%. Also, the area under ROC curve
(receiver operating characteristic curve) or AUC, which ranges between 0 and 1.0, is
employed to evaluate the learning performance of the machine. When AUC = 1, the
machine can correctly differentiate between all the positive and the negative class
points; however, when AUC = 0, the machine evaluates all negatives as positives, and
vice versa. Numerous DL models have been introduced for caries detection and their
performance has been investigated as well. A CNN-based system showed promising
accuracy (>80%) when detecting both the presence of caries and their location using
the oral photographs captured with digital cameras. Using another CNN-based
software (dentalXrai GmbH, Berlin, Germany), that employs bitewing radiographs,
significantly improved the accuracy of dentists for caries recognition especially for
enamel caries rather than dentin caries.
• Detection of tooth preparation
margins
Accurate determination of tooth preparation margin lines plays a crucial role in the
marginal fitness of dental restorations and especially indirect prostheses. In
conventional dentistry, the margin line is usually determined by using manual
interaction, which may be difficult, time-consuming, and complicated.
• Application of AI for detection of tooth preparation margins
In digital dentistry, margin line determination may be performed through three
methods of manual, semiautomatic, and automatic extraction. The manual
extraction is simple for the operator to manually choose feature points on the tooth
preparation model, while it can be less efficient. Regarding semiautomatic
extraction, the operator chooses multiple feature points sequentially on the path of
the tooth feature line while employing a search algorithm to instinctively extract the
margin line between the feature points. This technique is simple but inefficient. In
automatic extraction, the feature line is automatically extracted based on the
curvature and topological relationship by using differential geometry. This method is
more automated and efficient while minimizing manual interventions. Hence, AI and
DL models may be assets for mapping the preparation finish line in prosthodontics.
• Function of AI for detection of tooth preparation margins
The tooth preparation finish line can be automatically determined using CNN
models. The network model should be initially trained by using elaborated features
and images of tooth preparations. The more the number of tooth preparation
features, the more the accuracy of the network model. A study reported that the
robustness of a network model for extraction of the finish line could reach a high
level by the acquisition of 5000 tooth preparation samples. These samples can be
obtained virtually or physically; however, in both methods, the extracted finish line
may be a little bit inaccurate considering the actual finish line. This can be modified
by acquiring more tooth preparation and parametric adjustments.
• Accuracy of AI for detection of tooth preparation margins
One study estimated an accuracy of 97.43% for a newly-released model for finish
line extraction based on an S-Octree CNN model for three-dimensional shape
analysis. Two systematic reviews showed the promising abilities of AI models in
locating the tooth preparation finish line. However, owing to the limited numbers of
AI-based systems specified for finish line extraction and insufficient relevant
literature, making a generalization about the level of accuracy of AI for the detection
of tooth preparation margins is difficult and irrational, and therefore, more
investigations are required in this field.
• Detection of tooth restoration design
In conventional dentistry, dental reconstruction starts with the design of tooth
restorations which is usually performed by dental technicians manually. This
procedure models a tooth crown, a full tooth crown, multiple tooth crowns, or a full
mouth construction; meanwhile, it can be money, material, and time-consuming.
• Application of AI for tooth restoration design
Digital dentistry by using computer-aided design and computer-aided
manufacturing (CAD-CAM) systems benefits from dental software systems and
intraoral scanners to provide three-dimensional surface models of remaining teeth,
on which dental restorations are designed. The designed restorations can be milled
or printed in a final material. In this context, AI models can be used to automate the
design of dental restorations through customized reconstruction.
• Function of AI for tooth restoration design
For tooth restoration design, both the size and the shape of the teeth should be
reproduced.AI models can acquire and analyze the complicated shapes of
destructed teeth, neighboring and opposing teeth, and interocclusal relations
obtained from scanned data in order to reconstruct the surfaces of missing
structures. This process begins with the analysis of a wide range of tooth shapes
obtained from dental plaster casts available in dental labs or clinics. Thus, the
training data for ML are usually based on a large number of impressions taken of
intact and caries-free teeth that are cast with dental stones. Stone casts are usually
scanned by using three-dimensional scanners to create a large set of images needed
for ML. The tooth occlusal landmarks are defined and the distances between the
landmarks are measured to create the dataset needed for AI models. AI models
provide a biogeneric tooth reconstruction as an output for a scanned image of the
destructed tooth as an input.
• Accuracy of AI for tooth restoration design
A mean deviation of 150 μm has been reported for a biogeneric tooth model's
reconstruction from the original tooth surface. In a study, the crowns manually
reconstructed by dental technicians were less accurate than the crowns made by a
fully automatic software process regarding the original natural tooth as a control.
Even with all tooth cusps destruction, a biogeneric CAD software (CEREC v3.80,
Dentsply Sirona, York, PA) offered a more accurate tooth reconstruction than
performing wax-ups.
• Tooth shade determination
Tooth shade selection and reproduction are among the most difficult tasks in
esthetic and restorative dentistry. Creating color matches between dental
restorations and natural teeth is complicated because of the multi-dimensional
nature of color, various optical properties of natural teeth, and diverse influential
elements that can impact the resulting color.
• Application of AI for tooth shade determination
Various color-matching software programs have been released to help dental
clinicians and technicians with shade prediction and reproduction while targeting
objective modern instrumental methods rather than subjective traditional visual
methods. Computer-aided color-matching (CCM) has been used in the painting,
printing, and textile industries by providing instructional recipes to make a color
using the Kubelka–Munk theory. This theory shows how the color of a background is
changed by the use of a particular composition and thickness of a layer of paint
especially the thickness of paint required to mask the background. For example,
adding different specific color pigments to a white liquid can provide a specific color
for the liquid in the painting industry. Nevertheless, adding pigments to dental
restorative materials such as resins and ceramics cannot be simply performed
because of possible adverse effects on the material's properties and difficulties and
controls in manufacturing processes. To tackle this complexity and nonlinearity in
dentistry, ANNs, and BPNNs may be assets for modeling an enhancing system.
• Function of AI for tooth shade determination
A BPNN color-matching model consists of a multilayer FFNN along with back-
propagation algorithms. BPNN acts as an interactive gradient model to reduce color
mismatches between the predicted shade and the target shade. In backward
propagation, when the color mismatch exceeds an acceptable limit, the shade
determination process is back-propagated by tuning the weights and biases, as a
result, the process resumes till the color mismatch decreases to the acceptable
limit. For training a CCM system in dentistry by using BPNN, the color coordinates of
a set of specimens can be used as inputs, while their equivalent porcelain powder
recipes such as the weight of different types/layers of porcelain (body dentin and
enamel) serve as outputs. A combination of BPNN and genetic algorithms has been
proposed to improve the performance of CCM systems.
• Accuracy of AI for tooth shade determination
Similar to other AI models, a set of testing specimens with identified color
coordinates and porcelain powder recipes is employed to verify the accuracy of a
CCM model. Moreover, the color differences between the testing specimens and
their CCM-predicted outputs are calculated from color difference formulas. Then, the
color difference values are compared with clinical visual thresholds to clinically
evaluate the visibility of the color differences; In a study, a CCM system showed
acceptable color differences (less than the thresholds) when matching the colors of
natural teeth and zirconia-based restorations. Also, a BPNN-based CCM model
demonstrated superior shade reproduction outcomes compared with visual shade
selection methods while providing a desirable accuracy with color differences less
than the thresholds.
•Conclusion
Applications, functions, and accuracy of AI models for caries detection, tooth
preparation margin detection, tooth restoration design, metal structure casting,
dental restoration/implant detection, RPD design, and tooth shade determination
were discussed. Accordingly, a wide range of levels of accuracy have been reported
for the AI models discussed in the current review because of the several factors
impacting the accuracy such as the type of AI model, the source and size of training
data, the method of validation, and the control methods/groups. Based on the
current literature, the AI models have shown promising performance in the
mentioned aspects of restorative dentistry when being compared with traditional
approaches in terms of accuracy; however, as these AI models are still in
development, more studies are required to validate their accuracy and apply them to
routine clinical practice.

Artificial intelligence in restorative dentistry.pptx

  • 1.
    Student name:Ibrahim SaadMahmoud Subject:operative level:four
  • 2.
    ARTIFICIAL INTELLIGENCE INRESTORATIVE DENTISTRY
  • 3.
    • Introduction Artificial intelligence(AI) is a field of computer science that engages in rendering machines the ability to look like they have human intelligence or the capability to replicate intelligent human behaviors. In other words, AI is intelligence that is shown by machines, computers, and systems, whereas natural intelligence is represented by animals and humans. Accordingly, any machine that recognizes its surroundings and reacts to them for attaining specific objectives is considered as an intelligent agent. In this context, it is not surprising that the use of AI is significantly increasing in dentistry in which diagnosis, clinical decision-making, and prediction of treatment outcomes play fundamental roles. AI systems can help dentists with designing clinical decision support structures, and therefore, dentists should be aware of how AI works and how they can be assisted by AI systems.
  • 4.
    • Caries Detection Forpreserving tooth tissues, delaying restorative interventions, diminishing expensive restorative treatments, and retaining teeth long-term, early detection of non-cavitated lesions is a must for controlling these lesions through noninvasive treatments such as applying fluoride varnishes, infiltrating caries, and sealing fissures. The diagnosis of non-cavitated caries lesion with direct visual and tactile exploration may be inadequate for proximal tooth surfaces, and therefore, bitewing radiography is additionally suggested. However, the results of radiographic assessments for caries diagnosis may be affected by dental clinicians and examiners due to their false-positive or false-negative detections.
  • 5.
    • Application ofAI for caries detection AI models have been proposed as assistance to dental examiners when analyzing radiographic and/or photographic images in order to improve caries diagnosis in terms of accuracy and precision. AI models can employ intraoral digital images, bitewing radiographs, or panoramic radiographs to recognize caries lesions and offer appropriate treatments accordingly; however, panoramic radiographs may be less viable when it comes to caries detection.
  • 6.
    • Function ofAI for caries detection DL (deep learning) using CNNs(convolutional neural networks) has been mostly used to develop an AI-based system for caries detection. CNNs are able to map images as inputs and provide image identification and classification as outputs according to the weights obtained from training data. To train the AI-based system, images are initially labeled, and then, both the images and their labels are supplied to the CNN. The CNN is computationally programmed so that it can diagnose the presence of a labeled entity such as a caries lesion on unseen images. Indeed, CNNs are well- designed for image classification and are the most commonly used algorithms for caries identification.
  • 7.
    • Accuracy ofAI for caries detection When assessing the effectiveness of ML, the diagnostic accuracy of a machine is basically crucial and should be ideally 100%. Also, the area under ROC curve (receiver operating characteristic curve) or AUC, which ranges between 0 and 1.0, is employed to evaluate the learning performance of the machine. When AUC = 1, the machine can correctly differentiate between all the positive and the negative class points; however, when AUC = 0, the machine evaluates all negatives as positives, and vice versa. Numerous DL models have been introduced for caries detection and their performance has been investigated as well. A CNN-based system showed promising accuracy (>80%) when detecting both the presence of caries and their location using the oral photographs captured with digital cameras. Using another CNN-based software (dentalXrai GmbH, Berlin, Germany), that employs bitewing radiographs, significantly improved the accuracy of dentists for caries recognition especially for enamel caries rather than dentin caries.
  • 8.
    • Detection oftooth preparation margins Accurate determination of tooth preparation margin lines plays a crucial role in the marginal fitness of dental restorations and especially indirect prostheses. In conventional dentistry, the margin line is usually determined by using manual interaction, which may be difficult, time-consuming, and complicated.
  • 9.
    • Application ofAI for detection of tooth preparation margins In digital dentistry, margin line determination may be performed through three methods of manual, semiautomatic, and automatic extraction. The manual extraction is simple for the operator to manually choose feature points on the tooth preparation model, while it can be less efficient. Regarding semiautomatic extraction, the operator chooses multiple feature points sequentially on the path of the tooth feature line while employing a search algorithm to instinctively extract the margin line between the feature points. This technique is simple but inefficient. In automatic extraction, the feature line is automatically extracted based on the curvature and topological relationship by using differential geometry. This method is more automated and efficient while minimizing manual interventions. Hence, AI and DL models may be assets for mapping the preparation finish line in prosthodontics.
  • 10.
    • Function ofAI for detection of tooth preparation margins The tooth preparation finish line can be automatically determined using CNN models. The network model should be initially trained by using elaborated features and images of tooth preparations. The more the number of tooth preparation features, the more the accuracy of the network model. A study reported that the robustness of a network model for extraction of the finish line could reach a high level by the acquisition of 5000 tooth preparation samples. These samples can be obtained virtually or physically; however, in both methods, the extracted finish line may be a little bit inaccurate considering the actual finish line. This can be modified by acquiring more tooth preparation and parametric adjustments.
  • 11.
    • Accuracy ofAI for detection of tooth preparation margins One study estimated an accuracy of 97.43% for a newly-released model for finish line extraction based on an S-Octree CNN model for three-dimensional shape analysis. Two systematic reviews showed the promising abilities of AI models in locating the tooth preparation finish line. However, owing to the limited numbers of AI-based systems specified for finish line extraction and insufficient relevant literature, making a generalization about the level of accuracy of AI for the detection of tooth preparation margins is difficult and irrational, and therefore, more investigations are required in this field.
  • 12.
    • Detection oftooth restoration design In conventional dentistry, dental reconstruction starts with the design of tooth restorations which is usually performed by dental technicians manually. This procedure models a tooth crown, a full tooth crown, multiple tooth crowns, or a full mouth construction; meanwhile, it can be money, material, and time-consuming.
  • 13.
    • Application ofAI for tooth restoration design Digital dentistry by using computer-aided design and computer-aided manufacturing (CAD-CAM) systems benefits from dental software systems and intraoral scanners to provide three-dimensional surface models of remaining teeth, on which dental restorations are designed. The designed restorations can be milled or printed in a final material. In this context, AI models can be used to automate the design of dental restorations through customized reconstruction.
  • 14.
    • Function ofAI for tooth restoration design For tooth restoration design, both the size and the shape of the teeth should be reproduced.AI models can acquire and analyze the complicated shapes of destructed teeth, neighboring and opposing teeth, and interocclusal relations obtained from scanned data in order to reconstruct the surfaces of missing structures. This process begins with the analysis of a wide range of tooth shapes obtained from dental plaster casts available in dental labs or clinics. Thus, the training data for ML are usually based on a large number of impressions taken of intact and caries-free teeth that are cast with dental stones. Stone casts are usually scanned by using three-dimensional scanners to create a large set of images needed for ML. The tooth occlusal landmarks are defined and the distances between the landmarks are measured to create the dataset needed for AI models. AI models provide a biogeneric tooth reconstruction as an output for a scanned image of the destructed tooth as an input.
  • 15.
    • Accuracy ofAI for tooth restoration design A mean deviation of 150 μm has been reported for a biogeneric tooth model's reconstruction from the original tooth surface. In a study, the crowns manually reconstructed by dental technicians were less accurate than the crowns made by a fully automatic software process regarding the original natural tooth as a control. Even with all tooth cusps destruction, a biogeneric CAD software (CEREC v3.80, Dentsply Sirona, York, PA) offered a more accurate tooth reconstruction than performing wax-ups.
  • 16.
    • Tooth shadedetermination Tooth shade selection and reproduction are among the most difficult tasks in esthetic and restorative dentistry. Creating color matches between dental restorations and natural teeth is complicated because of the multi-dimensional nature of color, various optical properties of natural teeth, and diverse influential elements that can impact the resulting color.
  • 17.
    • Application ofAI for tooth shade determination Various color-matching software programs have been released to help dental clinicians and technicians with shade prediction and reproduction while targeting objective modern instrumental methods rather than subjective traditional visual methods. Computer-aided color-matching (CCM) has been used in the painting, printing, and textile industries by providing instructional recipes to make a color using the Kubelka–Munk theory. This theory shows how the color of a background is changed by the use of a particular composition and thickness of a layer of paint especially the thickness of paint required to mask the background. For example, adding different specific color pigments to a white liquid can provide a specific color for the liquid in the painting industry. Nevertheless, adding pigments to dental restorative materials such as resins and ceramics cannot be simply performed because of possible adverse effects on the material's properties and difficulties and controls in manufacturing processes. To tackle this complexity and nonlinearity in dentistry, ANNs, and BPNNs may be assets for modeling an enhancing system.
  • 18.
    • Function ofAI for tooth shade determination A BPNN color-matching model consists of a multilayer FFNN along with back- propagation algorithms. BPNN acts as an interactive gradient model to reduce color mismatches between the predicted shade and the target shade. In backward propagation, when the color mismatch exceeds an acceptable limit, the shade determination process is back-propagated by tuning the weights and biases, as a result, the process resumes till the color mismatch decreases to the acceptable limit. For training a CCM system in dentistry by using BPNN, the color coordinates of a set of specimens can be used as inputs, while their equivalent porcelain powder recipes such as the weight of different types/layers of porcelain (body dentin and enamel) serve as outputs. A combination of BPNN and genetic algorithms has been proposed to improve the performance of CCM systems.
  • 19.
    • Accuracy ofAI for tooth shade determination Similar to other AI models, a set of testing specimens with identified color coordinates and porcelain powder recipes is employed to verify the accuracy of a CCM model. Moreover, the color differences between the testing specimens and their CCM-predicted outputs are calculated from color difference formulas. Then, the color difference values are compared with clinical visual thresholds to clinically evaluate the visibility of the color differences; In a study, a CCM system showed acceptable color differences (less than the thresholds) when matching the colors of natural teeth and zirconia-based restorations. Also, a BPNN-based CCM model demonstrated superior shade reproduction outcomes compared with visual shade selection methods while providing a desirable accuracy with color differences less than the thresholds.
  • 20.
    •Conclusion Applications, functions, andaccuracy of AI models for caries detection, tooth preparation margin detection, tooth restoration design, metal structure casting, dental restoration/implant detection, RPD design, and tooth shade determination were discussed. Accordingly, a wide range of levels of accuracy have been reported for the AI models discussed in the current review because of the several factors impacting the accuracy such as the type of AI model, the source and size of training data, the method of validation, and the control methods/groups. Based on the current literature, the AI models have shown promising performance in the mentioned aspects of restorative dentistry when being compared with traditional approaches in terms of accuracy; however, as these AI models are still in development, more studies are required to validate their accuracy and apply them to routine clinical practice.